Re: [coreutils] testsuite failures on Fedora

[Jim Meyering]

Pádraig Brady wrote:

> On 17/09/10 11:06, Pádraig Brady wrote:
>> On 17/09/10 10:34, Jim Meyering wrote:
>>>> FAIL: misc/stat-mount (exit: 1)
>>>> ===============================
>>> ...
>>>> + stat_mnt=/shared/home
>>>> + test /shared = /shared/home
>>>> + fail=1
>>> ...
>>>> Here, the failure is that I have /home bind-mounted on /shared/home,
>>>> and since I ran the test under /home, the mount location reported by
>>>> df is different than the one reported by stat -c%m.  This is a new
>>>> test, and I'm not sure what to do about the issue (whether stat needs
>>>> fixing to deal with bind mounts, or whether the test is at fault).
>>>
>>> If there's a reliable mechanism to determine
>>> that there is a potentially interfering bind mount,
>>> I'd like to skip this test.
>>
>> Oops, invalid test.
>> I had forgotten I resolved the bind mount in that case also.
>> I'll cook up a fix this evening to skip the test if the inodes match
>> or something like that.
>
> Note `stat -c%m` was outputting the bind mount (/shared/home) in this case,
> while df was just outputting the original mount point for the device.
>
> One could assume the difference is due to bind mounts if
> the dev,inode were the same and pass in that case.
> Alternatively, one could run `stat -c%m` in a loop to get
> the base device mount point. However one could contrive a
> situation (using bind mounts) where df is outputting
> the original mount point for the device, while stat is
> outputting the current different alias.
> So in summary it's usually but not always valid
> to compare the mount points from `df` and `stat`.
> So I'll adjust the test as follows, to just
> check that stat outputs something.

Thanks.
This certainly won't provoke a false positive ;-)